@anvia/lancedb 1.0.0-rc.1 → 1.0.0-rc.10

package/dist/index.d.ts

@@ -1,60 +1,61 @@
-import { VectorFilter, VectorSearchIndex, VectorSearchRequest, VectorSearchResult, VectorSearchToolOptions } from '@anvia/core/vector-store';
-import { VectorMetadata, EmbeddingModel, EmbeddedDocument } from '@anvia/core/embeddings';
-import { Tool } from '@anvia/core/tool';
+import { VectorMetadata } from '@anvia/core/embeddings';
+import { VectorMetric, VectorStore, VectorStoreUpsertOptions, VectorSearchRequest, VectorSearchResult } from '@anvia/core/vector-store';
 
-declare function filterToLanceExpr(filter: VectorFilter | undefined): string | undefined;
-
-type LanceDBDistance = "cosine" | "l2" | "dot";
 type LanceDBTableLike = {
-    add(rows: Record<string, unknown>[]): Promise<unknown>;
-    search(vector: number[]): {
-        limit(n: number): {
-            filter(expr: string | undefined): Promise<unknown[]>;
-            toArray(): Promise<unknown[]>;
-        };
-    };
+    add(rows: Record<string, unknown>[], options?: Record<string, unknown>): Promise<unknown>;
+    search(vector: number[]): LanceDBQueryLike;
     countRows(): Promise<number>;
     delete(filter: string): Promise<unknown>;
+    schema?(): Promise<{
+        fields?: Array<{
+            name?: string;
+            type?: unknown;
+        }>;
+    }>;
+};
+type LanceDBQueryLike = {
+    limit(n: number): LanceDBQueryLike;
+    where(predicate: string): LanceDBQueryLike;
+    distanceType(metric: "cosine" | "l2" | "dot"): LanceDBQueryLike;
+    toArray(): Promise<unknown[]>;
 };
 type LanceDBConnectionLike = {
     openTable(name: string): Promise<LanceDBTableLike>;
     tableNames(): Promise<string[]>;
     createTable(name: string, data: Record<string, unknown>[]): Promise<LanceDBTableLike>;
+    createEmptyTable?(name: string, schema: unknown, options?: Record<string, unknown>): Promise<LanceDBTableLike>;
+    close?(): Promise<unknown> | unknown;
 };
-type LanceDBVectorStoreConnectOptions = {
+type LanceDBVectorClientOptions = {
     client?: LanceDBConnectionLike | undefined;
     uri?: string | undefined;
+};
+type LanceDBVectorStoreOptions = {
     tableName: string;
-    vectorSize: number;
-    createIfMissing?: boolean | undefined;
-    distance?: LanceDBDistance | undefined;
+    dimensions: number;
+    metric?: VectorMetric | undefined;
 };
 
-declare class LanceDBVectorIndex<T, Metadata extends VectorMetadata = VectorMetadata> implements VectorSearchIndex<T, Metadata> {
-    private readonly model;
-    private readonly connection;
-    private readonly tableName;
-    constructor(model: EmbeddingModel, connection: LanceDBConnectionLike, tableName: string);
+declare class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata> implements VectorStore<T, Metadata> {
+    private readonly owner;
+    readonly options: LanceDBVectorStoreOptions;
+    constructor(owner: LanceDBVectorClient, options: LanceDBVectorStoreOptions);
+    ensure(): Promise<void>;
+    validate(): Promise<void>;
+    upsert(options: VectorStoreUpsertOptions<T, Metadata>): Promise<void>;
     search(request: VectorSearchRequest): Promise<Array<VectorSearchResult<T, Metadata>>>;
-    searchIds(request: VectorSearchRequest): Promise<Array<{
-        score: number;
-        id: string;
-    }>>;
-    asTool(options: VectorSearchToolOptions): Tool<{
-        query: string;
-        topK?: number;
-    }, unknown>;
 }
 
-declare class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata> {
-    private readonly connection;
-    private readonly tableName;
-    private readonly distance;
-    private readonly vectorSize;
-    private constructor();
-    static connect<T, Metadata extends VectorMetadata = VectorMetadata>(options: LanceDBVectorStoreConnectOptions): Promise<LanceDBVectorStore<T, Metadata>>;
-    upsertDocuments(documents: Array<EmbeddedDocument<T, Metadata>>): Promise<void>;
-    index(model: EmbeddingModel): LanceDBVectorIndex<T, Metadata>;
+declare class LanceDBVectorClient {
+    private readonly options;
+    private readonly injected;
+    private clientPromise;
+    private closed;
+    constructor(options?: LanceDBVectorClientOptions);
+    vectorStore<T, Metadata extends VectorMetadata = VectorMetadata>(options: LanceDBVectorStoreOptions): LanceDBVectorStore<T, Metadata>;
+    nativeClient(): Promise<LanceDBConnectionLike>;
+    close(): Promise<void>;
+    private assertOpen;
 }
 
-export { type LanceDBConnectionLike, type LanceDBDistance, type LanceDBTableLike, LanceDBVectorIndex, LanceDBVectorStore, type LanceDBVectorStoreConnectOptions, filterToLanceExpr };
+export { type LanceDBConnectionLike, type LanceDBTableLike, LanceDBVectorClient, type LanceDBVectorClientOptions, LanceDBVectorStore, type LanceDBVectorStoreOptions };

package/dist/index.js

@@ -1,50 +1,12 @@
-// src/filters.ts
-function filterToLanceExpr(filter) {
-  if (filter === void 0) {
-    return void 0;
-  }
-  return translateFilter(filter);
-}
-function translateFilter(filter) {
-  switch (filter.type) {
-    case "eq":
-      if (typeof filter.value === "string") {
-        return `${filter.key} = '${escapeSql(filter.value)}'`;
-      }
-      if (typeof filter.value === "boolean") {
-        return `${filter.key} = ${filter.value ? "TRUE" : "FALSE"}`;
-      }
-      if (filter.value === null) {
-        return `${filter.key} IS NULL`;
-      }
-      return `${filter.key} = ${filter.value}`;
-    case "gt":
-      return `${filter.key} > ${filter.value}`;
-    case "lt":
-      return `${filter.key} < ${filter.value}`;
-    case "and":
-      return filter.filters.map((f) => `(${translateFilter(f)})`).join(" AND ");
-    case "or":
-      return filter.filters.map((f) => `(${translateFilter(f)})`).join(" OR ");
-  }
-}
-function escapeSql(value) {
-  return value.replace(/'/g, "''");
-}
-
-// src/search-index.ts
-import { embedText } from "@anvia/core/embeddings";
-import {
-  createVectorSearchTool
-} from "@anvia/core/vector-store";
-
 // src/helpers.ts
 import { createHash } from "crypto";
 
 // src/types.ts
 var documentIdColumn = "__anvia_document_id";
 var documentColumn = "__anvia_document";
+var metadataColumn = "__anvia_metadata";
 var vectorColumn = "__anvia_vector";
+var rowIdColumn = "__anvia_id";
 var reservedColumnPrefix = "__anvia_";
 
 // src/helpers.ts
@@ -82,22 +44,22 @@ function lanceRows(document) {
   return document.embeddings.map((embedding, index) => {
     const logicalId = document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;
     const row = {
-      _rowid: rowId(logicalId),
+      [rowIdColumn]: rowId(logicalId),
       [documentIdColumn]: document.id,
       [documentColumn]: serializeDocument(document.document),
+      [metadataColumn]: JSON.stringify(document.metadata ?? {}),
       [vectorColumn]: embedding.vector
     };
-    Object.assign(row, document.metadata);
     return row;
   });
 }
-function parseQueryResults(response, threshold) {
+function parseQueryResults(response, minScore, metric) {
   const byId = /* @__PURE__ */ new Map();
   for (const row of response) {
     const record = row;
     const distance = typeof record._distance === "number" ? record._distance : 0;
-    const score = 1 - distance;
-    if (threshold !== void 0 && score < threshold) {
+    const score = metric === void 0 || metric === "cosine" ? 1 - distance : -distance;
+    if (minScore !== void 0 && score < minScore) {
       continue;
     }
     const id = String(record[documentIdColumn] ?? "");
@@ -124,92 +86,180 @@ async function defaultLanceDBConnection(uri) {
   );
 }
 function metadataFromColumns(record) {
-  const metadata = Object.fromEntries(
-    Object.entries(record).filter(
-      ([key]) => !key.startsWith(reservedColumnPrefix) && key !== vectorColumn && key !== "_rowid" && key !== "_distance"
-    )
-  );
+  const raw = record[metadataColumn];
+  if (raw === null || raw === void 0) return void 0;
+  const metadata = typeof raw === "string" ? JSON.parse(raw) : raw;
   return Object.keys(metadata).length === 0 ? void 0 : metadata;
 }
 
-// src/search-index.ts
-var LanceDBVectorIndex = class {
-  constructor(model, connection, tableName) {
-    this.model = model;
-    this.connection = connection;
-    this.tableName = tableName;
-  }
-  model;
-  connection;
-  tableName;
-  async search(request) {
-    const queryEmbedding = await embedText(this.model, request.query);
-    const table = await this.connection.openTable(this.tableName);
-    const filterExpr = filterToLanceExpr(request.filter);
-    const query = table.search(queryEmbedding.vector).limit(request.topK);
-    let response;
-    if (filterExpr !== void 0) {
-      response = await query.filter(filterExpr);
-    } else {
-      response = await query.toArray();
+// src/store.ts
+import { matchesVectorFilter } from "@anvia/core/vector-store";
+var LanceDBVectorStore = class {
+  constructor(owner, options) {
+    this.owner = owner;
+    this.options = options;
+    assertDimensions(options.dimensions);
+  }
+  owner;
+  options;
+  async ensure() {
+    const connection = await this.owner.nativeClient();
+    if (!(await connection.tableNames()).includes(this.options.tableName)) {
+      if (connection.createEmptyTable === void 0) {
+        throw new TypeError("LanceDB provisioning requires createEmptyTable(...).");
+      }
+      const { Field, FixedSizeList, Float32, Schema, Utf8 } = await import("apache-arrow");
+      await connection.createEmptyTable(
+        this.options.tableName,
+        new Schema([
+          new Field(rowIdColumn, new Utf8(), false),
+          new Field(documentIdColumn, new Utf8(), false),
+          new Field(documentColumn, new Utf8(), false),
+          new Field(metadataColumn, new Utf8(), false),
+          new Field(
+            vectorColumn,
+            new FixedSizeList(this.options.dimensions, new Field("item", new Float32(), true)),
+            false
+          )
+        ])
+      );
     }
-    return parseQueryResults(response, request.threshold);
+    await this.validate();
   }
-  async searchIds(request) {
-    return (await this.search(request)).map(({ score, id }) => ({ score, id }));
+  async validate() {
+    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);
+    if (table.schema === void 0) return;
+    const schema = await table.schema();
+    const vectorField = schema.fields?.find((field) => field.name === vectorColumn);
+    if (vectorField === void 0) {
+      throw new Error(`LanceDB table ${this.options.tableName} is missing ${vectorColumn}`);
+    }
+    if (!schema.fields?.some((field) => field.name === metadataColumn)) {
+      throw new Error(`LanceDB table ${this.options.tableName} is missing ${metadataColumn}`);
+    }
+    const listSize = vectorField.type?.listSize;
+    if (typeof listSize === "number" && listSize !== this.options.dimensions) {
+      throw new Error(
+        `LanceDB table ${this.options.tableName} has ${listSize} dimensions; expected ${this.options.dimensions}`
+      );
+    }
   }
-  asTool(options) {
-    return createVectorSearchTool(this, options);
+  async upsert(options) {
+    validateDocuments(options.documents, this.options.dimensions);
+    if (options.documents.length === 0) return;
+    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);
+    const ids = options.documents.map((document) => sqlString(document.id)).join(", ");
+    await table.delete(`${documentIdColumn} IN (${ids})`);
+    const rows = options.documents.flatMap((document) => lanceRows(document));
+    if (rows.length > 0) await table.add(rows, options.providerOptions);
   }
-};
-
-// src/store.ts
-var LanceDBVectorStore = class _LanceDBVectorStore {
-  constructor(connection, tableName, distance, vectorSize) {
-    this.connection = connection;
-    this.tableName = tableName;
-    this.distance = distance;
-    this.vectorSize = vectorSize;
-  }
-  connection;
-  tableName;
-  distance;
-  vectorSize;
-  static async connect(options) {
-    const connection = options.client ?? await defaultLanceDBConnection(options.uri);
-    const distance = options.distance ?? "cosine";
-    if (options.createIfMissing === false) {
-      await connection.openTable(options.tableName);
-      return new _LanceDBVectorStore(
-        connection,
-        options.tableName,
-        distance,
-        options.vectorSize
-      );
+  async search(request) {
+    validateSearchRequest(request, this.options.dimensions);
+    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);
+    let candidateLimit = request.topK;
+    for (; ; ) {
+      const query = table.search(request.vector).distanceType(metricName(this.options.metric)).limit(candidateLimit);
+      const response = await query.toArray();
+      throwIfAborted(request.abortSignal);
+      const unfiltered = parseQueryResults(response, void 0, this.options.metric);
+      const results = selectResults(unfiltered, request);
+      if (results.length >= request.topK || response.length < candidateLimit || hasResultBelowMinimum(unfiltered, request.minScore))
+        return results;
+      const nextLimit = nextCandidateLimit(candidateLimit);
+      if (nextLimit === candidateLimit) return results;
+      candidateLimit = nextLimit;
     }
-    const tableNames = await connection.tableNames();
-    if (!tableNames.includes(options.tableName)) {
-      await connection.createTable(options.tableName, []);
+  }
+};
+function selectResults(results, request) {
+  return results.filter((result) => matchesVectorFilter(result.metadata, request.filter)).filter((result) => request.minScore === void 0 || result.score >= request.minScore).sort((left, right) => right.score - left.score).slice(0, request.topK);
+}
+function hasResultBelowMinimum(results, minScore) {
+  return minScore !== void 0 && results.some((result) => result.score < minScore);
+}
+function nextCandidateLimit(current) {
+  return current > Math.floor(Number.MAX_SAFE_INTEGER / 2) ? current : current * 2;
+}
+function sqlString(value) {
+  return `'${value.replaceAll("'", "''")}'`;
+}
+function throwIfAborted(signal) {
+  if (signal?.aborted) {
+    const error = new Error("The operation was aborted.");
+    error.name = "AbortError";
+    throw error;
+  }
+}
+function assertDimensions(dimensions) {
+  if (!Number.isSafeInteger(dimensions) || dimensions < 1)
+    throw new RangeError("Vector dimensions must be a positive safe integer.");
+}
+function validateDocuments(documents, dimensions) {
+  const ids = /* @__PURE__ */ new Set();
+  for (const document of documents) {
+    if (ids.has(document.id)) throw new TypeError(`Duplicate vector document id: ${document.id}`);
+    ids.add(document.id);
+    if (document.embeddings.length === 0)
+      throw new TypeError(`Vector document ${document.id} must contain at least one embedding.`);
+    for (const embedding of document.embeddings) {
+      if (embedding.vector.length !== dimensions)
+        throw new Error(`Vector dimension mismatch for document ${document.id}`);
+      if (!embedding.vector.every(Number.isFinite))
+        throw new Error(`Vector for document ${document.id} must contain only finite numbers`);
     }
-    return new _LanceDBVectorStore(
-      connection,
-      options.tableName,
-      distance,
-      options.vectorSize
+  }
+}
+function metricName(metric) {
+  if (metric === "euclidean") return "l2";
+  if (metric === "dotProduct") return "dot";
+  return "cosine";
+}
+function validateSearchRequest(request, dimensions) {
+  throwIfAborted(request.abortSignal);
+  if (!Number.isSafeInteger(request.topK) || request.topK < 1)
+    throw new RangeError("Vector search topK must be a positive safe integer.");
+  if (request.minScore !== void 0 && !Number.isFinite(request.minScore))
+    throw new RangeError("Vector search minScore must be a finite number.");
+  if (request.vector.length !== dimensions) {
+    throw new Error(
+      `Vector dimension mismatch: expected ${dimensions} dimensions but received ${request.vector.length} for query`
     );
   }
-  async upsertDocuments(documents) {
-    const table = await this.connection.openTable(this.tableName);
-    const rows = documents.flatMap((document) => lanceRows(document));
-    await table.add(rows);
+  if (!request.vector.every(Number.isFinite))
+    throw new Error("Vector search query must contain only finite numbers");
+}
+
+// src/client.ts
+var LanceDBVectorClient = class {
+  constructor(options = {}) {
+    this.options = options;
+    this.injected = options.client;
+  }
+  options;
+  injected;
+  clientPromise;
+  closed = false;
+  vectorStore(options) {
+    this.assertOpen();
+    return new LanceDBVectorStore(this, options);
+  }
+  nativeClient() {
+    this.assertOpen();
+    this.clientPromise ??= this.injected === void 0 ? defaultLanceDBConnection(this.options.uri) : Promise.resolve(this.injected);
+    return this.clientPromise;
+  }
+  async close() {
+    if (this.closed) return;
+    this.closed = true;
+    if (this.injected !== void 0 || this.clientPromise === void 0) return;
+    await (await this.clientPromise).close?.();
   }
-  index(model) {
-    return new LanceDBVectorIndex(model, this.connection, this.tableName);
+  assertOpen() {
+    if (this.closed) throw new Error("LanceDBVectorClient is closed.");
   }
 };
 export {
-  LanceDBVectorIndex,
-  LanceDBVectorStore,
-  filterToLanceExpr
+  LanceDBVectorClient,
+  LanceDBVectorStore
 };
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/filters.ts","../src/search-index.ts","../src/helpers.ts","../src/types.ts","../src/store.ts"],"sourcesContent":["import type { VectorFilter } from \"@anvia/core/vector-store\";\n\nexport function filterToLanceExpr(filter: VectorFilter | undefined): string | undefined {\n  if (filter === undefined) {\n    return undefined;\n  }\n\n  return translateFilter(filter);\n}\n\nfunction translateFilter(filter: VectorFilter): string {\n  switch (filter.type) {\n    case \"eq\":\n      if (typeof filter.value === \"string\") {\n        return `${filter.key} = '${escapeSql(filter.value)}'`;\n      }\n      if (typeof filter.value === \"boolean\") {\n        return `${filter.key} = ${filter.value ? \"TRUE\" : \"FALSE\"}`;\n      }\n      if (filter.value === null) {\n        return `${filter.key} IS NULL`;\n      }\n      return `${filter.key} = ${filter.value}`;\n    case \"gt\":\n      return `${filter.key} > ${filter.value}`;\n    case \"lt\":\n      return `${filter.key} < ${filter.value}`;\n    case \"and\":\n      return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" AND \");\n    case \"or\":\n      return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" OR \");\n  }\n}\n\nfunction escapeSql(value: string): string {\n  return value.replace(/'/g, \"''\");\n}\n","import { type EmbeddingModel, embedText, type VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { Tool } from \"@anvia/core/tool\";\nimport {\n  createVectorSearchTool,\n  type VectorSearchIndex,\n  type VectorSearchRequest,\n  type VectorSearchResult,\n  type VectorSearchToolOptions,\n} from \"@anvia/core/vector-store\";\nimport { filterToLanceExpr } from \"./filters.js\";\nimport { parseQueryResults } from \"./helpers.js\";\nimport type { LanceDBConnectionLike } from \"./types.js\";\n\nexport class LanceDBVectorIndex<T, Metadata extends VectorMetadata = VectorMetadata>\n  implements VectorSearchIndex<T, Metadata>\n{\n  constructor(\n    private readonly model: EmbeddingModel,\n    private readonly connection: LanceDBConnectionLike,\n    private readonly tableName: string,\n  ) {}\n\n  async search(request: VectorSearchRequest): Promise<Array<VectorSearchResult<T, Metadata>>> {\n    const queryEmbedding = await embedText(this.model, request.query);\n    const table = await this.connection.openTable(this.tableName);\n    const filterExpr = filterToLanceExpr(request.filter);\n\n    const query = table.search(queryEmbedding.vector).limit(request.topK);\n\n    let response: unknown[];\n    if (filterExpr !== undefined) {\n      response = (await query.filter(filterExpr)) as unknown[];\n    } else {\n      response = (await query.toArray()) as unknown[];\n    }\n\n    return parseQueryResults<T, Metadata>(response, request.threshold);\n  }\n\n  async searchIds(request: VectorSearchRequest): Promise<Array<{ score: number; id: string }>> {\n    return (await this.search(request)).map(({ score, id }) => ({ score, id }));\n  }\n\n  asTool(options: VectorSearchToolOptions): Tool<{ query: string; topK?: number }, unknown> {\n    return createVectorSearchTool(this, options);\n  }\n}\n","import { createHash } from \"node:crypto\";\nimport type { EmbeddedDocument, VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { VectorSearchResult } from \"@anvia/core/vector-store\";\nimport { documentColumn, documentIdColumn, reservedColumnPrefix, vectorColumn } from \"./types.js\";\n\nexport function assertNoReservedMetadata(metadata: VectorMetadata | undefined): void {\n  for (const key of Object.keys(metadata ?? {})) {\n    if (key.startsWith(reservedColumnPrefix)) {\n      throw new Error(`Metadata key ${key} is reserved for Anvia LanceDB columns`);\n    }\n  }\n}\n\nexport function rowId(id: string): string {\n  return createHash(\"sha256\").update(id).digest(\"hex\").slice(0, 32);\n}\n\nexport function serializeDocument(document: unknown): string {\n  return typeof document === \"string\" ? document : JSON.stringify(document);\n}\n\nexport function parseDocument<T>(document: unknown): T {\n  if (document === null || document === undefined) {\n    return \"\" as T;\n  }\n  if (typeof document !== \"string\") {\n    return document as T;\n  }\n  try {\n    return JSON.parse(document) as T;\n  } catch {\n    return document as T;\n  }\n}\n\nexport function lanceRows<T, Metadata extends VectorMetadata>(\n  document: EmbeddedDocument<T, Metadata>,\n): Array<Record<string, unknown>> {\n  if (document.embeddings.length === 0) {\n    throw new Error(`Document ${document.id} has no embeddings`);\n  }\n  assertNoReservedMetadata(document.metadata);\n\n  return document.embeddings.map((embedding, index) => {\n    const logicalId =\n      document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;\n    const row: Record<string, unknown> = {\n      _rowid: rowId(logicalId),\n      [documentIdColumn]: document.id,\n      [documentColumn]: serializeDocument(document.document),\n      [vectorColumn]: embedding.vector,\n    };\n    Object.assign(row, document.metadata);\n    return row;\n  });\n}\n\nexport function parseQueryResults<T, Metadata extends VectorMetadata>(\n  response: unknown[],\n  threshold: number | undefined,\n): Array<VectorSearchResult<T, Metadata>> {\n  const byId = new Map<string, VectorSearchResult<T, Metadata>>();\n\n  for (const row of response) {\n    const record = row as Record<string, unknown>;\n    const distance = typeof record._distance === \"number\" ? record._distance : 0;\n    const score = 1 - distance;\n\n    if (threshold !== undefined && score < threshold) {\n      continue;\n    }\n\n    const id = String(record[documentIdColumn] ?? \"\");\n    const result: VectorSearchResult<T, Metadata> = {\n      id,\n      score,\n      document: parseDocument(record[documentColumn]),\n    };\n    const metadata = metadataFromColumns<Metadata>(record);\n    if (metadata !== undefined) {\n      result.metadata = metadata;\n    }\n    const current = byId.get(id);\n    if (current === undefined || result.score > current.score) {\n      byId.set(id, result);\n    }\n  }\n\n  return [...byId.values()];\n}\n\nexport async function defaultLanceDBConnection(\n  uri?: string,\n): Promise<import(\"./types.js\").LanceDBConnectionLike> {\n  const lancedb = await import(\"@lancedb/lancedb\");\n  return lancedb.connect(\n    uri ?? \"~/.anvia/lancedb\",\n  ) as unknown as import(\"./types.js\").LanceDBConnectionLike;\n}\n\nfunction metadataFromColumns<Metadata extends VectorMetadata>(\n  record: Record<string, unknown>,\n): Metadata | undefined {\n  const metadata = Object.fromEntries(\n    Object.entries(record).filter(\n      ([key]) =>\n        !key.startsWith(reservedColumnPrefix) &&\n        key !== vectorColumn &&\n        key !== \"_rowid\" &&\n        key !== \"_distance\",\n    ),\n  ) as Metadata;\n  return Object.keys(metadata).length === 0 ? undefined : metadata;\n}\n","export const documentIdColumn = \"__anvia_document_id\";\nexport const documentColumn = \"__anvia_document\";\nexport const vectorColumn = \"__anvia_vector\";\nexport const reservedColumnPrefix = \"__anvia_\";\n\nexport type LanceDBDistance = \"cosine\" | \"l2\" | \"dot\";\n\nexport type LanceDBTableLike = {\n  add(rows: Record<string, unknown>[]): Promise<unknown>;\n  search(vector: number[]): {\n    limit(n: number): {\n      filter(expr: string | undefined): Promise<unknown[]>;\n      toArray(): Promise<unknown[]>;\n    };\n  };\n  countRows(): Promise<number>;\n  delete(filter: string): Promise<unknown>;\n};\n\nexport type LanceDBConnectionLike = {\n  openTable(name: string): Promise<LanceDBTableLike>;\n  tableNames(): Promise<string[]>;\n  createTable(name: string, data: Record<string, unknown>[]): Promise<LanceDBTableLike>;\n};\n\nexport type LanceDBVectorStoreConnectOptions = {\n  client?: LanceDBConnectionLike | undefined;\n  uri?: string | undefined;\n  tableName: string;\n  vectorSize: number;\n  createIfMissing?: boolean | undefined;\n  distance?: LanceDBDistance | undefined;\n};\n","import type { EmbeddedDocument, EmbeddingModel, VectorMetadata } from \"@anvia/core/embeddings\";\nimport { defaultLanceDBConnection, lanceRows } from \"./helpers.js\";\nimport { LanceDBVectorIndex } from \"./search-index.js\";\nimport type {\n  LanceDBConnectionLike,\n  LanceDBDistance,\n  LanceDBVectorStoreConnectOptions,\n} from \"./types.js\";\n\nexport class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata> {\n  private constructor(\n    private readonly connection: LanceDBConnectionLike,\n    private readonly tableName: string,\n    private readonly distance: LanceDBDistance,\n    private readonly vectorSize: number,\n  ) {}\n\n  static async connect<T, Metadata extends VectorMetadata = VectorMetadata>(\n    options: LanceDBVectorStoreConnectOptions,\n  ): Promise<LanceDBVectorStore<T, Metadata>> {\n    const connection = options.client ?? (await defaultLanceDBConnection(options.uri));\n    const distance = options.distance ?? \"cosine\";\n\n    if (options.createIfMissing === false) {\n      await connection.openTable(options.tableName);\n      return new LanceDBVectorStore<T, Metadata>(\n        connection,\n        options.tableName,\n        distance,\n        options.vectorSize,\n      );\n    }\n\n    const tableNames = await connection.tableNames();\n    if (!tableNames.includes(options.tableName)) {\n      await connection.createTable(options.tableName, []);\n    }\n\n    return new LanceDBVectorStore<T, Metadata>(\n      connection,\n      options.tableName,\n      distance,\n      options.vectorSize,\n    );\n  }\n\n  async upsertDocuments(documents: Array<EmbeddedDocument<T, Metadata>>): Promise<void> {\n    const table = await this.connection.openTable(this.tableName);\n    const rows = documents.flatMap((document) => lanceRows(document));\n    await table.add(rows);\n  }\n\n  index(model: EmbeddingModel): LanceDBVectorIndex<T, Metadata> {\n    return new LanceDBVectorIndex(model, this.connection, this.tableName);\n  }\n}\n"],"mappings":";AAEO,SAAS,kBAAkB,QAAsD;AACtF,MAAI,WAAW,QAAW;AACxB,WAAO;AAAA,EACT;AAEA,SAAO,gBAAgB,MAAM;AAC/B;AAEA,SAAS,gBAAgB,QAA8B;AACrD,UAAQ,OAAO,MAAM;AAAA,IACnB,KAAK;AACH,UAAI,OAAO,OAAO,UAAU,UAAU;AACpC,eAAO,GAAG,OAAO,GAAG,OAAO,UAAU,OAAO,KAAK,CAAC;AAAA,MACpD;AACA,UAAI,OAAO,OAAO,UAAU,WAAW;AACrC,eAAO,GAAG,OAAO,GAAG,MAAM,OAAO,QAAQ,SAAS,OAAO;AAAA,MAC3D;AACA,UAAI,OAAO,UAAU,MAAM;AACzB,eAAO,GAAG,OAAO,GAAG;AAAA,MACtB;AACA,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,OAAO;AAAA,IAC1E,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,MAAM;AAAA,EAC3E;AACF;AAEA,SAAS,UAAU,OAAuB;AACxC,SAAO,MAAM,QAAQ,MAAM,IAAI;AACjC;;;ACpCA,SAA8B,iBAAsC;AAEpE;AAAA,EACE;AAAA,OAKK;;;ACRP,SAAS,kBAAkB;;;ACApB,IAAM,mBAAmB;AACzB,IAAM,iBAAiB;AACvB,IAAM,eAAe;AACrB,IAAM,uBAAuB;;;ADE7B,SAAS,yBAAyB,UAA4C;AACnF,aAAW,OAAO,OAAO,KAAK,YAAY,CAAC,CAAC,GAAG;AAC7C,QAAI,IAAI,WAAW,oBAAoB,GAAG;AACxC,YAAM,IAAI,MAAM,gBAAgB,GAAG,wCAAwC;AAAA,IAC7E;AAAA,EACF;AACF;AAEO,SAAS,MAAM,IAAoB;AACxC,SAAO,WAAW,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AAClE;AAEO,SAAS,kBAAkB,UAA2B;AAC3D,SAAO,OAAO,aAAa,WAAW,WAAW,KAAK,UAAU,QAAQ;AAC1E;AAEO,SAAS,cAAiB,UAAsB;AACrD,MAAI,aAAa,QAAQ,aAAa,QAAW;AAC/C,WAAO;AAAA,EACT;AACA,MAAI,OAAO,aAAa,UAAU;AAChC,WAAO;AAAA,EACT;AACA,MAAI;AACF,WAAO,KAAK,MAAM,QAAQ;AAAA,EAC5B,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEO,SAAS,UACd,UACgC;AAChC,MAAI,SAAS,WAAW,WAAW,GAAG;AACpC,UAAM,IAAI,MAAM,YAAY,SAAS,EAAE,oBAAoB;AAAA,EAC7D;AACA,2BAAyB,SAAS,QAAQ;AAE1C,SAAO,SAAS,WAAW,IAAI,CAAC,WAAW,UAAU;AACnD,UAAM,YACJ,SAAS,WAAW,WAAW,IAAI,SAAS,KAAK,GAAG,SAAS,EAAE,cAAc,KAAK;AACpF,UAAM,MAA+B;AAAA,MACnC,QAAQ,MAAM,SAAS;AAAA,MACvB,CAAC,gBAAgB,GAAG,SAAS;AAAA,MAC7B,CAAC,cAAc,GAAG,kBAAkB,SAAS,QAAQ;AAAA,MACrD,CAAC,YAAY,GAAG,UAAU;AAAA,IAC5B;AACA,WAAO,OAAO,KAAK,SAAS,QAAQ;AACpC,WAAO;AAAA,EACT,CAAC;AACH;AAEO,SAAS,kBACd,UACA,WACwC;AACxC,QAAM,OAAO,oBAAI,IAA6C;AAE9D,aAAW,OAAO,UAAU;AAC1B,UAAM,SAAS;AACf,UAAM,WAAW,OAAO,OAAO,cAAc,WAAW,OAAO,YAAY;AAC3E,UAAM,QAAQ,IAAI;AAElB,QAAI,cAAc,UAAa,QAAQ,WAAW;AAChD;AAAA,IACF;AAEA,UAAM,KAAK,OAAO,OAAO,gBAAgB,KAAK,EAAE;AAChD,UAAM,SAA0C;AAAA,MAC9C;AAAA,MACA;AAAA,MACA,UAAU,cAAc,OAAO,cAAc,CAAC;AAAA,IAChD;AACA,UAAM,WAAW,oBAA8B,MAAM;AACrD,QAAI,aAAa,QAAW;AAC1B,aAAO,WAAW;AAAA,IACpB;AACA,UAAM,UAAU,KAAK,IAAI,EAAE;AAC3B,QAAI,YAAY,UAAa,OAAO,QAAQ,QAAQ,OAAO;AACzD,WAAK,IAAI,IAAI,MAAM;AAAA,IACrB;AAAA,EACF;AAEA,SAAO,CAAC,GAAG,KAAK,OAAO,CAAC;AAC1B;AAEA,eAAsB,yBACpB,KACqD;AACrD,QAAM,UAAU,MAAM,OAAO,kBAAkB;AAC/C,SAAO,QAAQ;AAAA,IACb,OAAO;AAAA,EACT;AACF;AAEA,SAAS,oBACP,QACsB;AACtB,QAAM,WAAW,OAAO;AAAA,IACtB,OAAO,QAAQ,MAAM,EAAE;AAAA,MACrB,CAAC,CAAC,GAAG,MACH,CAAC,IAAI,WAAW,oBAAoB,KACpC,QAAQ,gBACR,QAAQ,YACR,QAAQ;AAAA,IACZ;AAAA,EACF;AACA,SAAO,OAAO,KAAK,QAAQ,EAAE,WAAW,IAAI,SAAY;AAC1D;;;ADpGO,IAAM,qBAAN,MAEP;AAAA,EACE,YACmB,OACA,YACA,WACjB;AAHiB;AACA;AACA;AAAA,EAChB;AAAA,EAHgB;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,MAAM,OAAO,SAA+E;AAC1F,UAAM,iBAAiB,MAAM,UAAU,KAAK,OAAO,QAAQ,KAAK;AAChE,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,aAAa,kBAAkB,QAAQ,MAAM;AAEnD,UAAM,QAAQ,MAAM,OAAO,eAAe,MAAM,EAAE,MAAM,QAAQ,IAAI;AAEpE,QAAI;AACJ,QAAI,eAAe,QAAW;AAC5B,iBAAY,MAAM,MAAM,OAAO,UAAU;AAAA,IAC3C,OAAO;AACL,iBAAY,MAAM,MAAM,QAAQ;AAAA,IAClC;AAEA,WAAO,kBAA+B,UAAU,QAAQ,SAAS;AAAA,EACnE;AAAA,EAEA,MAAM,UAAU,SAA6E;AAC3F,YAAQ,MAAM,KAAK,OAAO,OAAO,GAAG,IAAI,CAAC,EAAE,OAAO,GAAG,OAAO,EAAE,OAAO,GAAG,EAAE;AAAA,EAC5E;AAAA,EAEA,OAAO,SAAmF;AACxF,WAAO,uBAAuB,MAAM,OAAO;AAAA,EAC7C;AACF;;;AGrCO,IAAM,qBAAN,MAAM,oBAAwE;AAAA,EAC3E,YACW,YACA,WACA,UACA,YACjB;AAJiB;AACA;AACA;AACA;AAAA,EAChB;AAAA,EAJgB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,aAAa,QACX,SAC0C;AAC1C,UAAM,aAAa,QAAQ,UAAW,MAAM,yBAAyB,QAAQ,GAAG;AAChF,UAAM,WAAW,QAAQ,YAAY;AAErC,QAAI,QAAQ,oBAAoB,OAAO;AACrC,YAAM,WAAW,UAAU,QAAQ,SAAS;AAC5C,aAAO,IAAI;AAAA,QACT;AAAA,QACA,QAAQ;AAAA,QACR;AAAA,QACA,QAAQ;AAAA,MACV;AAAA,IACF;AAEA,UAAM,aAAa,MAAM,WAAW,WAAW;AAC/C,QAAI,CAAC,WAAW,SAAS,QAAQ,SAAS,GAAG;AAC3C,YAAM,WAAW,YAAY,QAAQ,WAAW,CAAC,CAAC;AAAA,IACpD;AAEA,WAAO,IAAI;AAAA,MACT;AAAA,MACA,QAAQ;AAAA,MACR;AAAA,MACA,QAAQ;AAAA,IACV;AAAA,EACF;AAAA,EAEA,MAAM,gBAAgB,WAAgE;AACpF,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,OAAO,UAAU,QAAQ,CAAC,aAAa,UAAU,QAAQ,CAAC;AAChE,UAAM,MAAM,IAAI,IAAI;AAAA,EACtB;AAAA,EAEA,MAAM,OAAwD;AAC5D,WAAO,IAAI,mBAAmB,OAAO,KAAK,YAAY,KAAK,SAAS;AAAA,EACtE;AACF;","names":[]}
+{"version":3,"sources":["../src/helpers.ts","../src/types.ts","../src/store.ts","../src/client.ts"],"sourcesContent":["import { createHash } from \"node:crypto\";\nimport type { EmbeddedDocument, VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { VectorSearchResult } from \"@anvia/core/vector-store\";\nimport type { LanceDBVectorStoreOptions } from \"./types.js\";\nimport {\n  documentColumn,\n  documentIdColumn,\n  metadataColumn,\n  reservedColumnPrefix,\n  rowIdColumn,\n  vectorColumn,\n} from \"./types.js\";\n\nexport function assertNoReservedMetadata(metadata: VectorMetadata | undefined): void {\n  for (const key of Object.keys(metadata ?? {})) {\n    if (key.startsWith(reservedColumnPrefix)) {\n      throw new Error(`Metadata key ${key} is reserved for Anvia LanceDB columns`);\n    }\n  }\n}\n\nexport function rowId(id: string): string {\n  return createHash(\"sha256\").update(id).digest(\"hex\").slice(0, 32);\n}\n\nexport function serializeDocument(document: unknown): string {\n  return typeof document === \"string\" ? document : JSON.stringify(document);\n}\n\nexport function parseDocument<T>(document: unknown): T {\n  if (document === null || document === undefined) {\n    return \"\" as T;\n  }\n  if (typeof document !== \"string\") {\n    return document as T;\n  }\n  try {\n    return JSON.parse(document) as T;\n  } catch {\n    return document as T;\n  }\n}\n\nexport function lanceRows<T, Metadata extends VectorMetadata>(\n  document: EmbeddedDocument<T, Metadata>,\n): Array<Record<string, unknown>> {\n  if (document.embeddings.length === 0) {\n    throw new Error(`Document ${document.id} has no embeddings`);\n  }\n  assertNoReservedMetadata(document.metadata);\n\n  return document.embeddings.map((embedding, index) => {\n    const logicalId =\n      document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;\n    const row: Record<string, unknown> = {\n      [rowIdColumn]: rowId(logicalId),\n      [documentIdColumn]: document.id,\n      [documentColumn]: serializeDocument(document.document),\n      [metadataColumn]: JSON.stringify(document.metadata ?? {}),\n      [vectorColumn]: embedding.vector,\n    };\n    return row;\n  });\n}\n\nexport function parseQueryResults<T, Metadata extends VectorMetadata>(\n  response: unknown[],\n  minScore: number | undefined,\n  metric: LanceDBVectorStoreOptions[\"metric\"],\n): Array<VectorSearchResult<T, Metadata>> {\n  const byId = new Map<string, VectorSearchResult<T, Metadata>>();\n\n  for (const row of response) {\n    const record = row as Record<string, unknown>;\n    const distance = typeof record._distance === \"number\" ? record._distance : 0;\n    const score = metric === undefined || metric === \"cosine\" ? 1 - distance : -distance;\n\n    if (minScore !== undefined && score < minScore) {\n      continue;\n    }\n\n    const id = String(record[documentIdColumn] ?? \"\");\n    const result: VectorSearchResult<T, Metadata> = {\n      id,\n      score,\n      document: parseDocument(record[documentColumn]),\n    };\n    const metadata = metadataFromColumns<Metadata>(record);\n    if (metadata !== undefined) {\n      result.metadata = metadata;\n    }\n    const current = byId.get(id);\n    if (current === undefined || result.score > current.score) {\n      byId.set(id, result);\n    }\n  }\n\n  return [...byId.values()];\n}\n\nexport async function defaultLanceDBConnection(\n  uri?: string,\n): Promise<import(\"./types.js\").LanceDBConnectionLike> {\n  const lancedb = await import(\"@lancedb/lancedb\");\n  return lancedb.connect(\n    uri ?? \"~/.anvia/lancedb\",\n  ) as unknown as import(\"./types.js\").LanceDBConnectionLike;\n}\n\nfunction metadataFromColumns<Metadata extends VectorMetadata>(\n  record: Record<string, unknown>,\n): Metadata | undefined {\n  const raw = record[metadataColumn];\n  if (raw === null || raw === undefined) return undefined;\n  const metadata = (typeof raw === \"string\" ? JSON.parse(raw) : raw) as Metadata;\n  return Object.keys(metadata).length === 0 ? undefined : metadata;\n}\n","import type { VectorMetric } from \"@anvia/core/vector-store\";\n\nexport const documentIdColumn = \"__anvia_document_id\";\nexport const documentColumn = \"__anvia_document\";\nexport const metadataColumn = \"__anvia_metadata\";\nexport const vectorColumn = \"__anvia_vector\";\nexport const rowIdColumn = \"__anvia_id\";\nexport const reservedColumnPrefix = \"__anvia_\";\n\nexport type LanceDBTableLike = {\n  add(rows: Record<string, unknown>[], options?: Record<string, unknown>): Promise<unknown>;\n  search(vector: number[]): LanceDBQueryLike;\n  countRows(): Promise<number>;\n  delete(filter: string): Promise<unknown>;\n  schema?(): Promise<{ fields?: Array<{ name?: string; type?: unknown }> }>;\n};\n\nexport type LanceDBQueryLike = {\n  limit(n: number): LanceDBQueryLike;\n  where(predicate: string): LanceDBQueryLike;\n  distanceType(metric: \"cosine\" | \"l2\" | \"dot\"): LanceDBQueryLike;\n  toArray(): Promise<unknown[]>;\n};\n\nexport type LanceDBConnectionLike = {\n  openTable(name: string): Promise<LanceDBTableLike>;\n  tableNames(): Promise<string[]>;\n  createTable(name: string, data: Record<string, unknown>[]): Promise<LanceDBTableLike>;\n  createEmptyTable?(\n    name: string,\n    schema: unknown,\n    options?: Record<string, unknown>,\n  ): Promise<LanceDBTableLike>;\n  close?(): Promise<unknown> | unknown;\n};\n\nexport type LanceDBVectorClientOptions = {\n  client?: LanceDBConnectionLike | undefined;\n  uri?: string | undefined;\n};\n\nexport type LanceDBVectorStoreOptions = {\n  tableName: string;\n  dimensions: number;\n  metric?: VectorMetric | undefined;\n};\n","import type { EmbeddedDocument, VectorMetadata } from \"@anvia/core/embeddings\";\nimport type {\n  VectorSearchRequest,\n  VectorSearchResult,\n  VectorStore,\n  VectorStoreUpsertOptions,\n} from \"@anvia/core/vector-store\";\nimport { matchesVectorFilter } from \"@anvia/core/vector-store\";\nimport type { LanceDBVectorClient } from \"./client.js\";\nimport { lanceRows, parseQueryResults } from \"./helpers.js\";\nimport {\n  documentColumn,\n  documentIdColumn,\n  type LanceDBVectorStoreOptions,\n  metadataColumn,\n  rowIdColumn,\n  vectorColumn,\n} from \"./types.js\";\n\nexport class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata>\n  implements VectorStore<T, Metadata>\n{\n  constructor(\n    private readonly owner: LanceDBVectorClient,\n    readonly options: LanceDBVectorStoreOptions,\n  ) {\n    assertDimensions(options.dimensions);\n  }\n\n  async ensure(): Promise<void> {\n    const connection = await this.owner.nativeClient();\n    if (!(await connection.tableNames()).includes(this.options.tableName)) {\n      if (connection.createEmptyTable === undefined) {\n        throw new TypeError(\"LanceDB provisioning requires createEmptyTable(...).\");\n      }\n      const { Field, FixedSizeList, Float32, Schema, Utf8 } = await import(\"apache-arrow\");\n      await connection.createEmptyTable(\n        this.options.tableName,\n        new Schema([\n          new Field(rowIdColumn, new Utf8(), false),\n          new Field(documentIdColumn, new Utf8(), false),\n          new Field(documentColumn, new Utf8(), false),\n          new Field(metadataColumn, new Utf8(), false),\n          new Field(\n            vectorColumn,\n            new FixedSizeList(this.options.dimensions, new Field(\"item\", new Float32(), true)),\n            false,\n          ),\n        ]),\n      );\n    }\n    await this.validate();\n  }\n\n  async validate(): Promise<void> {\n    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);\n    if (table.schema === undefined) return;\n    const schema = await table.schema();\n    const vectorField = schema.fields?.find((field) => field.name === vectorColumn);\n    if (vectorField === undefined) {\n      throw new Error(`LanceDB table ${this.options.tableName} is missing ${vectorColumn}`);\n    }\n    if (!schema.fields?.some((field) => field.name === metadataColumn)) {\n      throw new Error(`LanceDB table ${this.options.tableName} is missing ${metadataColumn}`);\n    }\n    const listSize = (vectorField.type as { listSize?: unknown } | undefined)?.listSize;\n    if (typeof listSize === \"number\" && listSize !== this.options.dimensions) {\n      throw new Error(\n        `LanceDB table ${this.options.tableName} has ${listSize} dimensions; expected ${this.options.dimensions}`,\n      );\n    }\n  }\n\n  async upsert(options: VectorStoreUpsertOptions<T, Metadata>): Promise<void> {\n    validateDocuments(options.documents, this.options.dimensions);\n    if (options.documents.length === 0) return;\n    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);\n    const ids = options.documents.map((document) => sqlString(document.id)).join(\", \");\n    await table.delete(`${documentIdColumn} IN (${ids})`);\n    const rows = options.documents.flatMap((document) => lanceRows(document));\n    if (rows.length > 0) await table.add(rows, options.providerOptions);\n  }\n\n  async search(request: VectorSearchRequest): Promise<Array<VectorSearchResult<T, Metadata>>> {\n    validateSearchRequest(request, this.options.dimensions);\n    const table = await (await this.owner.nativeClient()).openTable(this.options.tableName);\n    let candidateLimit = request.topK;\n    for (;;) {\n      const query = table\n        .search(request.vector)\n        .distanceType(metricName(this.options.metric))\n        .limit(candidateLimit);\n      const response = await query.toArray();\n      throwIfAborted(request.abortSignal);\n      const unfiltered = parseQueryResults<T, Metadata>(response, undefined, this.options.metric);\n      const results = selectResults(unfiltered, request);\n      if (\n        results.length >= request.topK ||\n        response.length < candidateLimit ||\n        hasResultBelowMinimum(unfiltered, request.minScore)\n      )\n        return results;\n      const nextLimit = nextCandidateLimit(candidateLimit);\n      if (nextLimit === candidateLimit) return results;\n      candidateLimit = nextLimit;\n    }\n  }\n}\n\nfunction selectResults<T, Metadata extends VectorMetadata>(\n  results: Array<VectorSearchResult<T, Metadata>>,\n  request: VectorSearchRequest,\n): Array<VectorSearchResult<T, Metadata>> {\n  return results\n    .filter((result) => matchesVectorFilter(result.metadata, request.filter))\n    .filter((result) => request.minScore === undefined || result.score >= request.minScore)\n    .sort((left, right) => right.score - left.score)\n    .slice(0, request.topK);\n}\n\nfunction hasResultBelowMinimum<T, Metadata extends VectorMetadata>(\n  results: Array<VectorSearchResult<T, Metadata>>,\n  minScore: number | undefined,\n): boolean {\n  return minScore !== undefined && results.some((result) => result.score < minScore);\n}\n\nfunction nextCandidateLimit(current: number): number {\n  return current > Math.floor(Number.MAX_SAFE_INTEGER / 2) ? current : current * 2;\n}\n\nfunction sqlString(value: string): string {\n  return `'${value.replaceAll(\"'\", \"''\")}'`;\n}\n\nfunction throwIfAborted(signal: AbortSignal | undefined): void {\n  if (signal?.aborted) {\n    const error = new Error(\"The operation was aborted.\");\n    error.name = \"AbortError\";\n    throw error;\n  }\n}\n\nfunction assertDimensions(dimensions: number): void {\n  if (!Number.isSafeInteger(dimensions) || dimensions < 1)\n    throw new RangeError(\"Vector dimensions must be a positive safe integer.\");\n}\n\nfunction validateDocuments<T, Metadata extends VectorMetadata>(\n  documents: Array<EmbeddedDocument<T, Metadata>>,\n  dimensions: number,\n): void {\n  const ids = new Set<string>();\n  for (const document of documents) {\n    if (ids.has(document.id)) throw new TypeError(`Duplicate vector document id: ${document.id}`);\n    ids.add(document.id);\n    if (document.embeddings.length === 0)\n      throw new TypeError(`Vector document ${document.id} must contain at least one embedding.`);\n    for (const embedding of document.embeddings) {\n      if (embedding.vector.length !== dimensions)\n        throw new Error(`Vector dimension mismatch for document ${document.id}`);\n      if (!embedding.vector.every(Number.isFinite))\n        throw new Error(`Vector for document ${document.id} must contain only finite numbers`);\n    }\n  }\n}\n\nfunction metricName(metric: LanceDBVectorStoreOptions[\"metric\"]): \"cosine\" | \"l2\" | \"dot\" {\n  if (metric === \"euclidean\") return \"l2\";\n  if (metric === \"dotProduct\") return \"dot\";\n  return \"cosine\";\n}\n\nfunction validateSearchRequest(request: VectorSearchRequest, dimensions: number): void {\n  throwIfAborted(request.abortSignal);\n  if (!Number.isSafeInteger(request.topK) || request.topK < 1)\n    throw new RangeError(\"Vector search topK must be a positive safe integer.\");\n  if (request.minScore !== undefined && !Number.isFinite(request.minScore))\n    throw new RangeError(\"Vector search minScore must be a finite number.\");\n  if (request.vector.length !== dimensions) {\n    throw new Error(\n      `Vector dimension mismatch: expected ${dimensions} dimensions but received ${request.vector.length} for query`,\n    );\n  }\n  if (!request.vector.every(Number.isFinite))\n    throw new Error(\"Vector search query must contain only finite numbers\");\n}\n","import type { VectorMetadata } from \"@anvia/core/embeddings\";\nimport { defaultLanceDBConnection } from \"./helpers.js\";\nimport { LanceDBVectorStore } from \"./store.js\";\nimport type {\n  LanceDBConnectionLike,\n  LanceDBVectorClientOptions,\n  LanceDBVectorStoreOptions,\n} from \"./types.js\";\n\nexport class LanceDBVectorClient {\n  private readonly injected: LanceDBConnectionLike | undefined;\n  private clientPromise: Promise<LanceDBConnectionLike> | undefined;\n  private closed = false;\n\n  constructor(private readonly options: LanceDBVectorClientOptions = {}) {\n    this.injected = options.client;\n  }\n\n  vectorStore<T, Metadata extends VectorMetadata = VectorMetadata>(\n    options: LanceDBVectorStoreOptions,\n  ): LanceDBVectorStore<T, Metadata> {\n    this.assertOpen();\n    return new LanceDBVectorStore<T, Metadata>(this, options);\n  }\n\n  nativeClient(): Promise<LanceDBConnectionLike> {\n    this.assertOpen();\n    this.clientPromise ??=\n      this.injected === undefined\n        ? defaultLanceDBConnection(this.options.uri)\n        : Promise.resolve(this.injected);\n    return this.clientPromise;\n  }\n\n  async close(): Promise<void> {\n    if (this.closed) return;\n    this.closed = true;\n    if (this.injected !== undefined || this.clientPromise === undefined) return;\n    await (await this.clientPromise).close?.();\n  }\n\n  private assertOpen(): void {\n    if (this.closed) throw new Error(\"LanceDBVectorClient is closed.\");\n  }\n}\n"],"mappings":";AAAA,SAAS,kBAAkB;;;ACEpB,IAAM,mBAAmB;AACzB,IAAM,iBAAiB;AACvB,IAAM,iBAAiB;AACvB,IAAM,eAAe;AACrB,IAAM,cAAc;AACpB,IAAM,uBAAuB;;;ADM7B,SAAS,yBAAyB,UAA4C;AACnF,aAAW,OAAO,OAAO,KAAK,YAAY,CAAC,CAAC,GAAG;AAC7C,QAAI,IAAI,WAAW,oBAAoB,GAAG;AACxC,YAAM,IAAI,MAAM,gBAAgB,GAAG,wCAAwC;AAAA,IAC7E;AAAA,EACF;AACF;AAEO,SAAS,MAAM,IAAoB;AACxC,SAAO,WAAW,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AAClE;AAEO,SAAS,kBAAkB,UAA2B;AAC3D,SAAO,OAAO,aAAa,WAAW,WAAW,KAAK,UAAU,QAAQ;AAC1E;AAEO,SAAS,cAAiB,UAAsB;AACrD,MAAI,aAAa,QAAQ,aAAa,QAAW;AAC/C,WAAO;AAAA,EACT;AACA,MAAI,OAAO,aAAa,UAAU;AAChC,WAAO;AAAA,EACT;AACA,MAAI;AACF,WAAO,KAAK,MAAM,QAAQ;AAAA,EAC5B,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEO,SAAS,UACd,UACgC;AAChC,MAAI,SAAS,WAAW,WAAW,GAAG;AACpC,UAAM,IAAI,MAAM,YAAY,SAAS,EAAE,oBAAoB;AAAA,EAC7D;AACA,2BAAyB,SAAS,QAAQ;AAE1C,SAAO,SAAS,WAAW,IAAI,CAAC,WAAW,UAAU;AACnD,UAAM,YACJ,SAAS,WAAW,WAAW,IAAI,SAAS,KAAK,GAAG,SAAS,EAAE,cAAc,KAAK;AACpF,UAAM,MAA+B;AAAA,MACnC,CAAC,WAAW,GAAG,MAAM,SAAS;AAAA,MAC9B,CAAC,gBAAgB,GAAG,SAAS;AAAA,MAC7B,CAAC,cAAc,GAAG,kBAAkB,SAAS,QAAQ;AAAA,MACrD,CAAC,cAAc,GAAG,KAAK,UAAU,SAAS,YAAY,CAAC,CAAC;AAAA,MACxD,CAAC,YAAY,GAAG,UAAU;AAAA,IAC5B;AACA,WAAO;AAAA,EACT,CAAC;AACH;AAEO,SAAS,kBACd,UACA,UACA,QACwC;AACxC,QAAM,OAAO,oBAAI,IAA6C;AAE9D,aAAW,OAAO,UAAU;AAC1B,UAAM,SAAS;AACf,UAAM,WAAW,OAAO,OAAO,cAAc,WAAW,OAAO,YAAY;AAC3E,UAAM,QAAQ,WAAW,UAAa,WAAW,WAAW,IAAI,WAAW,CAAC;AAE5E,QAAI,aAAa,UAAa,QAAQ,UAAU;AAC9C;AAAA,IACF;AAEA,UAAM,KAAK,OAAO,OAAO,gBAAgB,KAAK,EAAE;AAChD,UAAM,SAA0C;AAAA,MAC9C;AAAA,MACA;AAAA,MACA,UAAU,cAAc,OAAO,cAAc,CAAC;AAAA,IAChD;AACA,UAAM,WAAW,oBAA8B,MAAM;AACrD,QAAI,aAAa,QAAW;AAC1B,aAAO,WAAW;AAAA,IACpB;AACA,UAAM,UAAU,KAAK,IAAI,EAAE;AAC3B,QAAI,YAAY,UAAa,OAAO,QAAQ,QAAQ,OAAO;AACzD,WAAK,IAAI,IAAI,MAAM;AAAA,IACrB;AAAA,EACF;AAEA,SAAO,CAAC,GAAG,KAAK,OAAO,CAAC;AAC1B;AAEA,eAAsB,yBACpB,KACqD;AACrD,QAAM,UAAU,MAAM,OAAO,kBAAkB;AAC/C,SAAO,QAAQ;AAAA,IACb,OAAO;AAAA,EACT;AACF;AAEA,SAAS,oBACP,QACsB;AACtB,QAAM,MAAM,OAAO,cAAc;AACjC,MAAI,QAAQ,QAAQ,QAAQ,OAAW,QAAO;AAC9C,QAAM,WAAY,OAAO,QAAQ,WAAW,KAAK,MAAM,GAAG,IAAI;AAC9D,SAAO,OAAO,KAAK,QAAQ,EAAE,WAAW,IAAI,SAAY;AAC1D;;;AE7GA,SAAS,2BAA2B;AAY7B,IAAM,qBAAN,MAEP;AAAA,EACE,YACmB,OACR,SACT;AAFiB;AACR;AAET,qBAAiB,QAAQ,UAAU;AAAA,EACrC;AAAA,EAJmB;AAAA,EACR;AAAA,EAKX,MAAM,SAAwB;AAC5B,UAAM,aAAa,MAAM,KAAK,MAAM,aAAa;AACjD,QAAI,EAAE,MAAM,WAAW,WAAW,GAAG,SAAS,KAAK,QAAQ,SAAS,GAAG;AACrE,UAAI,WAAW,qBAAqB,QAAW;AAC7C,cAAM,IAAI,UAAU,sDAAsD;AAAA,MAC5E;AACA,YAAM,EAAE,OAAO,eAAe,SAAS,QAAQ,KAAK,IAAI,MAAM,OAAO,cAAc;AACnF,YAAM,WAAW;AAAA,QACf,KAAK,QAAQ;AAAA,QACb,IAAI,OAAO;AAAA,UACT,IAAI,MAAM,aAAa,IAAI,KAAK,GAAG,KAAK;AAAA,UACxC,IAAI,MAAM,kBAAkB,IAAI,KAAK,GAAG,KAAK;AAAA,UAC7C,IAAI,MAAM,gBAAgB,IAAI,KAAK,GAAG,KAAK;AAAA,UAC3C,IAAI,MAAM,gBAAgB,IAAI,KAAK,GAAG,KAAK;AAAA,UAC3C,IAAI;AAAA,YACF;AAAA,YACA,IAAI,cAAc,KAAK,QAAQ,YAAY,IAAI,MAAM,QAAQ,IAAI,QAAQ,GAAG,IAAI,CAAC;AAAA,YACjF;AAAA,UACF;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AACA,UAAM,KAAK,SAAS;AAAA,EACtB;AAAA,EAEA,MAAM,WAA0B;AAC9B,UAAM,QAAQ,OAAO,MAAM,KAAK,MAAM,aAAa,GAAG,UAAU,KAAK,QAAQ,SAAS;AACtF,QAAI,MAAM,WAAW,OAAW;AAChC,UAAM,SAAS,MAAM,MAAM,OAAO;AAClC,UAAM,cAAc,OAAO,QAAQ,KAAK,CAAC,UAAU,MAAM,SAAS,YAAY;AAC9E,QAAI,gBAAgB,QAAW;AAC7B,YAAM,IAAI,MAAM,iBAAiB,KAAK,QAAQ,SAAS,eAAe,YAAY,EAAE;AAAA,IACtF;AACA,QAAI,CAAC,OAAO,QAAQ,KAAK,CAAC,UAAU,MAAM,SAAS,cAAc,GAAG;AAClE,YAAM,IAAI,MAAM,iBAAiB,KAAK,QAAQ,SAAS,eAAe,cAAc,EAAE;AAAA,IACxF;AACA,UAAM,WAAY,YAAY,MAA6C;AAC3E,QAAI,OAAO,aAAa,YAAY,aAAa,KAAK,QAAQ,YAAY;AACxE,YAAM,IAAI;AAAA,QACR,iBAAiB,KAAK,QAAQ,SAAS,QAAQ,QAAQ,yBAAyB,KAAK,QAAQ,UAAU;AAAA,MACzG;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,SAA+D;AAC1E,sBAAkB,QAAQ,WAAW,KAAK,QAAQ,UAAU;AAC5D,QAAI,QAAQ,UAAU,WAAW,EAAG;AACpC,UAAM,QAAQ,OAAO,MAAM,KAAK,MAAM,aAAa,GAAG,UAAU,KAAK,QAAQ,SAAS;AACtF,UAAM,MAAM,QAAQ,UAAU,IAAI,CAAC,aAAa,UAAU,SAAS,EAAE,CAAC,EAAE,KAAK,IAAI;AACjF,UAAM,MAAM,OAAO,GAAG,gBAAgB,QAAQ,GAAG,GAAG;AACpD,UAAM,OAAO,QAAQ,UAAU,QAAQ,CAAC,aAAa,UAAU,QAAQ,CAAC;AACxE,QAAI,KAAK,SAAS,EAAG,OAAM,MAAM,IAAI,MAAM,QAAQ,eAAe;AAAA,EACpE;AAAA,EAEA,MAAM,OAAO,SAA+E;AAC1F,0BAAsB,SAAS,KAAK,QAAQ,UAAU;AACtD,UAAM,QAAQ,OAAO,MAAM,KAAK,MAAM,aAAa,GAAG,UAAU,KAAK,QAAQ,SAAS;AACtF,QAAI,iBAAiB,QAAQ;AAC7B,eAAS;AACP,YAAM,QAAQ,MACX,OAAO,QAAQ,MAAM,EACrB,aAAa,WAAW,KAAK,QAAQ,MAAM,CAAC,EAC5C,MAAM,cAAc;AACvB,YAAM,WAAW,MAAM,MAAM,QAAQ;AACrC,qBAAe,QAAQ,WAAW;AAClC,YAAM,aAAa,kBAA+B,UAAU,QAAW,KAAK,QAAQ,MAAM;AAC1F,YAAM,UAAU,cAAc,YAAY,OAAO;AACjD,UACE,QAAQ,UAAU,QAAQ,QAC1B,SAAS,SAAS,kBAClB,sBAAsB,YAAY,QAAQ,QAAQ;AAElD,eAAO;AACT,YAAM,YAAY,mBAAmB,cAAc;AACnD,UAAI,cAAc,eAAgB,QAAO;AACzC,uBAAiB;AAAA,IACnB;AAAA,EACF;AACF;AAEA,SAAS,cACP,SACA,SACwC;AACxC,SAAO,QACJ,OAAO,CAAC,WAAW,oBAAoB,OAAO,UAAU,QAAQ,MAAM,CAAC,EACvE,OAAO,CAAC,WAAW,QAAQ,aAAa,UAAa,OAAO,SAAS,QAAQ,QAAQ,EACrF,KAAK,CAAC,MAAM,UAAU,MAAM,QAAQ,KAAK,KAAK,EAC9C,MAAM,GAAG,QAAQ,IAAI;AAC1B;AAEA,SAAS,sBACP,SACA,UACS;AACT,SAAO,aAAa,UAAa,QAAQ,KAAK,CAAC,WAAW,OAAO,QAAQ,QAAQ;AACnF;AAEA,SAAS,mBAAmB,SAAyB;AACnD,SAAO,UAAU,KAAK,MAAM,OAAO,mBAAmB,CAAC,IAAI,UAAU,UAAU;AACjF;AAEA,SAAS,UAAU,OAAuB;AACxC,SAAO,IAAI,MAAM,WAAW,KAAK,IAAI,CAAC;AACxC;AAEA,SAAS,eAAe,QAAuC;AAC7D,MAAI,QAAQ,SAAS;AACnB,UAAM,QAAQ,IAAI,MAAM,4BAA4B;AACpD,UAAM,OAAO;AACb,UAAM;AAAA,EACR;AACF;AAEA,SAAS,iBAAiB,YAA0B;AAClD,MAAI,CAAC,OAAO,cAAc,UAAU,KAAK,aAAa;AACpD,UAAM,IAAI,WAAW,oDAAoD;AAC7E;AAEA,SAAS,kBACP,WACA,YACM;AACN,QAAM,MAAM,oBAAI,IAAY;AAC5B,aAAW,YAAY,WAAW;AAChC,QAAI,IAAI,IAAI,SAAS,EAAE,EAAG,OAAM,IAAI,UAAU,iCAAiC,SAAS,EAAE,EAAE;AAC5F,QAAI,IAAI,SAAS,EAAE;AACnB,QAAI,SAAS,WAAW,WAAW;AACjC,YAAM,IAAI,UAAU,mBAAmB,SAAS,EAAE,uCAAuC;AAC3F,eAAW,aAAa,SAAS,YAAY;AAC3C,UAAI,UAAU,OAAO,WAAW;AAC9B,cAAM,IAAI,MAAM,0CAA0C,SAAS,EAAE,EAAE;AACzE,UAAI,CAAC,UAAU,OAAO,MAAM,OAAO,QAAQ;AACzC,cAAM,IAAI,MAAM,uBAAuB,SAAS,EAAE,mCAAmC;AAAA,IACzF;AAAA,EACF;AACF;AAEA,SAAS,WAAW,QAAsE;AACxF,MAAI,WAAW,YAAa,QAAO;AACnC,MAAI,WAAW,aAAc,QAAO;AACpC,SAAO;AACT;AAEA,SAAS,sBAAsB,SAA8B,YAA0B;AACrF,iBAAe,QAAQ,WAAW;AAClC,MAAI,CAAC,OAAO,cAAc,QAAQ,IAAI,KAAK,QAAQ,OAAO;AACxD,UAAM,IAAI,WAAW,qDAAqD;AAC5E,MAAI,QAAQ,aAAa,UAAa,CAAC,OAAO,SAAS,QAAQ,QAAQ;AACrE,UAAM,IAAI,WAAW,iDAAiD;AACxE,MAAI,QAAQ,OAAO,WAAW,YAAY;AACxC,UAAM,IAAI;AAAA,MACR,uCAAuC,UAAU,4BAA4B,QAAQ,OAAO,MAAM;AAAA,IACpG;AAAA,EACF;AACA,MAAI,CAAC,QAAQ,OAAO,MAAM,OAAO,QAAQ;AACvC,UAAM,IAAI,MAAM,sDAAsD;AAC1E;;;ACjLO,IAAM,sBAAN,MAA0B;AAAA,EAK/B,YAA6B,UAAsC,CAAC,GAAG;AAA1C;AAC3B,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAF6B;AAAA,EAJZ;AAAA,EACT;AAAA,EACA,SAAS;AAAA,EAMjB,YACE,SACiC;AACjC,SAAK,WAAW;AAChB,WAAO,IAAI,mBAAgC,MAAM,OAAO;AAAA,EAC1D;AAAA,EAEA,eAA+C;AAC7C,SAAK,WAAW;AAChB,SAAK,kBACH,KAAK,aAAa,SACd,yBAAyB,KAAK,QAAQ,GAAG,IACzC,QAAQ,QAAQ,KAAK,QAAQ;AACnC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,QAAuB;AAC3B,QAAI,KAAK,OAAQ;AACjB,SAAK,SAAS;AACd,QAAI,KAAK,aAAa,UAAa,KAAK,kBAAkB,OAAW;AACrE,WAAO,MAAM,KAAK,eAAe,QAAQ;AAAA,EAC3C;AAAA,EAEQ,aAAmB;AACzB,QAAI,KAAK,OAAQ,OAAM,IAAI,MAAM,gCAAgC;AAAA,EACnE;AACF;","names":[]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anvia/lancedb",
-  "version": "1.0.0-rc.1",
+  "version": "1.0.0-rc.10",
   "description": "LanceDB vector store adapter for Anvia.",
   "author": "anvia",
   "maintainer": "Indra Zulfi",
@@ -26,17 +26,18 @@
     }
   },
   "dependencies": {
-    "@lancedb/lancedb": "^0.33.0"
+    "@lancedb/lancedb": "^0.37.1",
+    "apache-arrow": "21.2.0"
   },
   "devDependencies": {
     "@types/node": "^24.9.1",
     "tsup": "^8.5.0",
     "typescript": "^5.9.3",
     "vitest": "^4.0.8",
-    "@anvia/core": "1.0.0-rc.1"
+    "@anvia/core": "1.0.0-rc.10"
   },
   "peerDependencies": {
-    "@anvia/core": "1.0.0-rc.1"
+    "@anvia/core": "1.0.0-rc.10"
   },
   "scripts": {
     "build": "tsup src/index.ts --format esm --dts --sourcemap --clean",